Balanced audio output : opamp configuration

[ratatax]

Hello

For the sound generator I design I have a PCM5102A DAC from Texas Instruments which outputs the sound, however it's a bit weak for line-level and i'd like to have the output as a balanced TRS jack (complementary/reversed signals) for use in studio conditions.

So we decided to use opamps to do the trick. I had someone giving me this schematic but he hadn't explained much about it, so as a developer (NOT an electronical engineer!) I tried to understand it and while I think it would work, I don't get the point of its relative complexity compared to the goal. I don't get the point of the first inverting amp, and why the final signal inversion is done with opamps in series instead of parallel.

If I follow the schematic it seems to do the job : the Tip of the jack will receive the original (inverted two times) amplified signal, while the Ring will get the reversed signal (inverted three times).

I feel it could be simplified with only two opamps instead of four : inject the DAC signal in two parallel opamps which both provide a gain, one in non-inverting mode the other in inverting mode.
However maybe I'm in a noob trap here, -- I struggle a bit understanding impedance-related things, maybe i'm missing something.

What do you guys think about it ?

[darrellg]

For the sound generator I design I have a PCM5102A DAC from Texas Instruments which outputs the sound, however it's a bit weak for line-level and i'd like to have the output as a balanced TRS jack (complementary/reversed signals) for use in studio conditions.

That's not what balanced means. That's the definition for differential signaling. A balanced signal pair is a transmission line consisting of two conductors of the same type, each of which have equal impedances along their lengths and equal impedances to ground and to other circuits. You can have a balanced line without a differential signal.

[DaJMasta]

Two parallel amps may be fine, but unless you're using matched resistors or very tight tolerance ones, their gains will be slightly different basically always, so keeping it single ended then converting to differential minimizes the independent drift/variation of two channels somewhat.


If it were me, I'd lay it out differently, there seems to be a fair bit of extra parts in this design.  First stage would be a unity gain inverting amp (like the output stages right now) but with the potentiometer in series with the feedback resistor, then a small capacitor in parallel with the pair of feedback resistors to take out high frequency noise above the audio band.  That would give you plenty of gain, and depending on the input impedance of your target device, you could either have a single inverting unity gain amp to give you the inverse signal and just the output of the gain stage to drive the negative differential output directly (half the number of opamps, many fewer parts), or, if you need to be able to source some current, you could have an inverting unity gain amp and a follower amp after the gain stage (3 opamps, but now each differential output stage can drive the full current rated for the opamp).  If running a single supply instead of a split supply, then you'd also want a cap on the output to remove the DC bias, and the resistor on the output probably doesn't need to be so high unless you want full protection from a constant short on the output.

[oPossum]

http://www.thatcorp.com/1600-series_Balanced_Line_Driver_ICs.shtml

[Someone]

There is a fairly comprehensive coverage of the topic here:
http://sound.whsites.net/articles/balanced-io.htm

[langwadt]

For the sound generator I design I have a PCM5102A DAC from Texas Instruments which outputs the sound, however it's a bit weak for line-level and i'd like to have the output as a balanced TRS jack (complementary/reversed signals) for use in studio conditions.

That's not what balanced means. That's the definition for differential signaling. A balanced signal pair is a transmission line consisting of two conductors of the same type, each of which have equal impedances along their lengths and equal impedances to ground and to other circuits. You can have a balanced line without a differential signal.

and in audio differential is commonly called balanced

[Bassman59]

For the sound generator I design I have a PCM5102A DAC from Texas Instruments which outputs the sound, however it's a bit weak for line-level and i'd like to have the output as a balanced TRS jack (complementary/reversed signals) for use in studio conditions.

That's not what balanced means. That's the definition for differential signaling. A balanced signal pair is a transmission line consisting of two conductors of the same type, each of which have equal impedances along their lengths and equal impedances to ground and to other circuits. You can have a balanced line without a differential signal.

and in audio differential is commonly called balanced

And they still use the word "phase" when they mean "polarity."

[bson]

Don't cascade the bridge halves, just pair an inverting and non-inverting amplifier.

[richard.cs]

It can also be done with a single opamp driving a transistor phase splitter. Opamp output to base, equal resistors down from the emitter and up from the collector and take the feedback for the opamp from the emitter. The opamp servos the voltage on the emitter resistor to be correct, and as the current is (nearly) the same in the collector resistor its voltage is the same just phase inverted. It gives reasonable matching between the complementary outputs but doesn't have the low and matched output impedance of the twin opamp solution.

[magic]

Opamp output to base, equal resistors down from the emitter and up from the collector and take the feedback for the opamp from the emitter.

BJT will give bad distortion from current gain variation with collector current and collector voltage. Maybe a FET would have some chance.
The typical solution in audio is simply what bson showed.
A diff-in, diff-out opamp like OPA1632 would also do.

[richard.cs]

BJT will give bad distortion from current gain variation with collector current and collector voltage. Maybe a FET would have some chance.

The current gain variation is (mostly) removed by the opamp feedback because that's taken from the emitter. All that's important is that the gain is large so that collector and emitter current are approximately equal - finite gain affects the balance of the output. There is some distortion on the  inverted (collector) signal because of the gain variation, but it can be small, especially if you start with a high gain device and set the circuit up with a high standing current such that the gain variation is small. It depends what your application is as to whether that matters.

I see no reason it won't work with a FET so long as you isolate the gate capacitance from the opamp so it doesn't oscillate.

[Bassman59]

BJT will give bad distortion from current gain variation with collector current and collector voltage. Maybe a FET would have some chance.

The current gain variation is (mostly) removed by the opamp feedback because that's taken from the emitter. All that's important is that the gain is large so that collector and emitter current are approximately equal - finite gain affects the balance of the output. There is some distortion on the  inverted (collector) signal because of the gain variation, but it can be small, especially if you start with a high gain device and set the circuit up with a high standing current such that the gain variation is small. It depends what your application is as to whether that matters.

I see no reason it won't work with a FET so long as you isolate the gate capacitance from the opamp so it doesn't oscillate.

The simplest way to do this correctly is to use a THAT1600-series or TI DRV134/135 part and be done with it.

[MagicSmoker]

...
If I follow the schematic it seems to do the job : the Tip of the jack will receive the original (inverted two times) amplified signal, while the Ring will get the reversed signal (inverted three times).

I feel it could be simplified with only two opamps instead of four : inject the DAC signal in two parallel opamps which both provide a gain, one in non-inverting mode the other in inverting mode.
...

I don't think anyone actually answered your question... The reason it is better to use two inverting amplifiers in series to produce a balanced (ie - differential) output is because it is more difficult to match the gain and input impedance of an inverting amplifier with a non-inverting (especially for unity gain versions of each). Any mismatch in gain will be treated as a legitimate differential signal by the downstream balanced receiver, after all.

[bson]

Small variations in gain won't matter.  The total gain is the difference between the two; Vo = K1*Vin - K2*Vin so you get Vo = (K1-K2)*Vin.   In other words, if K1 = 1.05 and K2 = -0.9 you get a total gain of 1.95 between them.  No problems here.

For an actual circuit you'll want a resistor equal to R1//R2 to ground on the positive input of the inverting amplifier, to balance the input offset currents, but that's not strictly necessary just to get it to work.

[magic]

I will add that the gain of opamp voltage follower is 0.99999 or so, you won't get the same level accuracy with resistors.
An inverting stage with 1% resistors may have up to 2% gain error. Two such stages - up to 4% gain mismatch.

[ratatax]

Thanks all for your answers ! I understand it a bit more now... maybe i'll keep the four opamp stages (just tweaking the values if needed after testing) since while sourcing the components I've noticed that that the quad version isn't much more expensive compared to the dual (like $ 0.2 vs 0.25) so we won't gain much by optimizing it except if we remove half of the opamps, which doesn't seems to be possible without affecting precision and balance if I understand correctly.

[chris_leyson]

LTC1992 is another option if you want to run everything from a single 5V supply. Rail to rail input and output with fixed gain of 1, 2, 5 and 10 or adjustable gain.

[Marco]

There is a fairly comprehensive coverage of the topic here:
http://sound.whsites.net/articles/balanced-io.htm

I think it's misleading that he says the ground compensated circuit has no common mode rejection, it's more accurate to say it introduces next to no common mode error to begin with.

It will blow everything else out of the water in that regard.

[David Hess]

Don't cascade the bridge halves, just pair an inverting and non-inverting amplifier.

The reason two cascaded inverters are actually not that bad is that an inverter and a follower have different noise gain and inverters have less distortion than followers which suffer from common mode effects.  An added difficulty is that your inverter and follower example presents different impedances to the operational amplifiers further compromising differential performance.

Walter Jung gives the higher performance example shown first below which cancels out the differences and the common mode distortion.  Note however that this is not a true differential output although it is suitable in most cases.

The second example is cross coupled for a true differential output and what THAT does.  Another example is shown here:

https://www.rane.com/note124.html

[Marco]

Does that ground compensated circuit actually work? When I simulate it, I actually get vastly increased coupled signals ... just leaving the ground feedback out (by deleting R2) improves matters immensely assuming a differential input on the other end (ie. measuring differential voltage across R10).

Am I missing something?

On another note, what's even the point of a true differential output? These lines are not properly terminated and even if they were the wavelengths at audio frequencies make keeping currents anti-symmetrical irrelevant, you don't have a fully differential signal chain which you want to keep fully differential either ... so what's the point? Keeping impedances symmetrical is all that's necessary AFAICS, so a common mode signal impressed on the signal pair stays common mode (a ground loop current is also a common mode signal). Tie one end to ground with a given resistance, the other end to a single ended amplifier with the same resistance, done ... why not?

[Bassman59]

Does that ground compensated circuit actually work? When I simulate it, I actually get vastly increased coupled signals ... just leaving the ground feedback out (by deleting R2) improves matters immensely assuming a differential input on the other end (ie. measuring differential voltage across R10).

Am I missing something?

Yes, it does. It certainly worked on the older Soundcraft and other British consoles from the early 80s.

On another note, what's even the point of a true differential output?

Double the output level (6 dB gain) for free. Well, at the cost of a second driver.

These lines are not properly terminated

Audio lines want to see a bridging load impedance, which is why the source driver impedance is ~50 ohms and the load side is ~10 k ohms. You're not dealing with RF so power transfer is not interesting and therefore you don't match impedances. (There's no such thing as a 600-ohm line in modern professional audio.) The point of the bridging impedance is that it minimizes the voltage-divider effect.  So there's no "termination" as one might think about in the RF or digital-signal world.

  Keeping impedances symmetrical is all that's necessary AFAICS, so a common mode signal impressed on the signal pair stays common mode (a ground loop current is also a common mode signal). Tie one end to ground with a given resistance, the other end to a single ended amplifier with the same resistance, done ... why not?

What you describe is what is done on pretty much every mixing console and piece of analog gear for the last, oh, 25 years, when they want to save a few pennies. It's got the unfortunate (and redundant) name "impedance balanced." To separate the lower-tier products from the higher tier, only the latter will have fully-differential outputs, everything else is impedance balanced. As a marketing bullet point, the manufacturer may want to advertise that the output can swing +26 dBu, which you can really only get with differential drive. Not that it matters a whole lot when everything is calibrated for +4 dBu "nominal," but headroom is nice and so is the ability to drive stupidly-long lines from FOH to the amp racks and still have some signal left.

[Marco]

So what am I missing? A common mode signal impressed on the lines with identical impedance for the ground balanced circuit gets forced into a non common mode signal AFAICS.

[David Hess]

The true differential output circuits like the two cross coupled examples I gave reject common mode changes even before the receiver gets involved.  If one of the outputs is shorted to ground for example, the differential output remains the same instead of being halved.  So asymmetrical effects on the differential pair are rejected by the transmitter.

[Jan Audio]

Why would you need the inverted sound ?
@bson : he need to have more gain, use a non inverting amplifier and some amplification for the inverting one.